Object-Based Time-Constrained Dynamic Time Warping Classification of Crops Using Sentinel-2

Round 1

Reviewer 1 Report

General comments

This manuscript applied an object-based multi-band time-constraint dynamic time wrapping approach in crop classification using Sentinel-2 time series data. The authors discussed the implications of DTW dissimilarity values in understanding the classification errors and evaluate the error propagations throughout our analysis. The authors also claimed that one important work is that they discussed the possible implications of DTW as a measure of spatio-temporal autocorrelation for SITS. However, the content of the discussion is limited, and no experiment supports this conclusion.

Overall, this study is interesting, and the manuscript is well organized. The manuscript can be accepted after some questions are addressed and detailed information is added.

Detailed comments:

L122: Please add a flowchart somewhere in Section 2

L196: How did you determine if they are representative or not? Are the reference and validation samples generated from the CropScape dataset?

L197: The authors generated NDVI temporal pattern as input, but how about for multiband DTW method?

L207-208: “For simplicity…” Maybe it is better to add another paragraph to describe the difference between different vegetation indices.

L305-L306: I believe the smaller the value, the similar the two sequences, but it is better to clearly state it here? Also stating the rule of classification using dissimilarity value here can help the readers to understand this algorithm.

Eq (3): Please explain the meaning of these parameters.

L313: “One of …”->Two of …?

L343: A flowchart would be better.

L381: How did you get the accuracy form CropScape? Does the dataset provide the accuracy?

L387: A figure of segmentation result would be better.

Figure 8: Maybe it is better to normalize the dissimilarity values for single-band and multi-band DTW to make the range of two are the same? Otherwise, how did you class the dissimilarity values in the figures since the ranges are not the same?

L489-L490: Not clear. (c) single band, DTW45 and (d) multi-band DTW30?

L545: Add subtitles.

L550: “while slightly decreasing it for Texas” why? Does it mean the multi-band DTW is not superior to sing-band DTW?

Author Response

Please find the pdf attached with a point-by-point response to the reviewer's comments, highlighted in red.

Author Response File: Author Response.pdf

Reviewer 2 Report

Object-Based Time-Constrained Dynamic Time Warping Classification of Crop Using Sentinel-2

Overall Comments

Image analysts are always looking for good “cookbook” techniques to use for processing remotely-sensed imagery.  The objective of this study was to compare multiple single-band and multi-band object-based time-constrained dynamic time warping classifications for crop mapping based on Sentinel-2 time series of vegetation indices. Another initiative of the study was to create a technique that an analyst could employ using the eCognition software.  The goals of the study were accomplished. 

The abstract provided a good summary of the study.  The Introduction was concise and clearly defined the purpose of the study. Job well done on the literature cited.  Overall, appropriate methods were used to obtain the results.  However, more detail is needed in a few sections of the Materials and Methods (see below). To improve the quality of the Results and Discussion sections, modifications are needed for the Figures and Tables---cosmetic changes; see comments below.     

Specific Comments

Abstract

1.      Page 1 of 26, line 31: change OBIA to object-based image analysis and then place OBIA in parenthesis

Introduction

1.      Page 2 of 26, line 56. Insert meaning of TIMESAT

Materials and Methods

1.      Insert the year used for CropScape data.

2.      Sentinel images:  Bands 3, 4, and 8 spatial resolution is 10m; bands 5, 6, 8A, and 11 spatial resolution is 20 m.  I assuming that resampling was done so that the column and rows of the 20 m data matched the number of columns and rows for the 10 m data? Insert that information into the manuscript. 

3.      Page 12 of 26, lines 381-383: More information is needed on how the data were altered. 

Results

1.      Section 3.3 User’s and Producer’s Accuracies

a.       A good explanation has been provided pertaining to errors in the classification. To make the section more informative show the actual error matrix table for the single band and multiband classifications summarized in Tables 6 and 7. 

References

1.      Common errors: (1) consistency in capitalization of journal article title; for some articles all the important words first letter appear in caps; other articles only the first word appears in caps, (2) spell out or abbreviate journal names, and (3) punctuation errors.

Figures

1.      The figures show the appropriate results. Nevertheless, figures should stand alone without the reader referring to the text for the meaning of abbreviations. Therefore, spell out meaning of acronyms the first time mentioned, followed by their shortened version in parenthesis.  Then use the abbreviation for the remainder of the figure caption. Applies to Figures 2-10.  Below are other comments pertaining to the figures.

a.       Figures 3 and 4.  Insert sub-figures in alphabetical order: Alfalfa 1, Alfalfa 2,…..That change would make it easier to find and read the graphs. Change Wheat to Winter Wheat. 

b.      Figure 5. DTW is time flexible; please provide a better explanation or example of its flexibility---flexible in what way. 

c.       Figures 9, 10. Only display the bottom half of the matrix; all items below the 1 diagonal---no need to show the perfect 1 diagonal. It appears that the top and bottom half of the matrix express the same results.     

Tables

1.      The tables contain the appropriate results. However, they should stand alone without the reader referring to the text for the meaning of abbreviations. Therefore, spell out meaning of acronyms. Note for tables the meanings can be added as a footnote below the table.   Applies to Tables 4, 5(DTW), 6(DTW), 7(DTW), 8, and 9.  Below are other comments pertaining to the figures.

2.      Table 5. In the text, percent is used as the unit value for overall accuracy. Therefore, insert percent as the unit value for the overall accuracies presented in Table 5, 75.2%. Another option would be to place the percent sign below the column header DTW0 (%).  Also, indicate in table caption the meaning of values shown in bold text.     

Author Response

Please find the pdf attached with a point-by-point response to the reviewer's comments, highlighted in red.

Author Response File: Author Response.pdf